Gemini 3.7 Flash

https://storage.googleapis.com/gweb-uniblog-publish-prod/images/gemini-3-7-flash__evals__perform.width-1200.format-webp.webp
Gemini 3.7 Flash is the latest model, offering substantial improvements in coding and agent tasks with a lower price point than Gemini 3.6 Flash. It delivers strong gains in debugging, issue resolution, and code accuracy across various workflows and domains.

DeepSeek Harness developer preview

https://deepseek.com/harness/images/harness/feat-plugin.en.png
DeepSeek Harness is now in developer preview, offering a plugin-based system for agent developers to create custom harnesses. It provides various modes, including standard, code, minimal and creator modes with tools and capabilities that can be swapped or extended.

Spaghettifying DRAM

https://raw.githubusercontent.com/xoreaxeaxeax/skitter-creek-bath-salts/main/examples/unspaghettify.gif
The skitter-creek-bath-salts project exploits a vulnerability in AMD Family 16h CPUs by manipulating the DRAM controller to scramble memory, bypassing security features like SEV and SGX. This allows for unrestricted access to protected regions of memory, including PSP private memory and SMRAM.

Codex in ChatGPT desktop app for Linux is now in preview

https://us1.discourse-cdn.com/openai1/original/4X/0/4/c/04c8546833f0824a49d21f50e2b08dd44bfd3d05.png
The ChatGPT desktop app for Linux is now available in preview, offering a native experience with supported Ubuntu and Debian versions. It provides a workspace for managing projects and running Codex alongside ChatGPT.

Accelerating GPT-5.6 Sol Ultrafast

https://cdn.sanity.io/images/e4qjo92p/production/96417454c2234a5f3241876a06719db0d27fae1e-1273x910.png?auto=format&dpr=2&fit=max&q=75&w=1273
Cerebras and OpenAI are launching Ultrafast Mode, a new service tier powered by Cerebras that delivers up to 750 output tokens per second without quality compromise. This resolves the tradeoff between speed and intelligence in AI models, enabling faster inference for mission-critical work.

Gloomberb

https://gloom.sh/_next/image?url=%2Flanding-terminal.png&w=3840&q=75
See what's new in 0.10.4 β–žβ–€β––β–œ β–Œ β–Œ β–Œβ–„β––β– β–žβ–€β––β–žβ–€β––β–›β–šβ–€β––β–›β–€β––β–žβ–€β––β–™β–€β––β–›β–€β–– β–Œ β–Œβ– β–Œ β–Œβ–Œ β–Œβ–Œβ– β–Œβ–Œ β–Œβ–›β–€ β–Œ β–Œ β–Œ ▝▀ β–˜β–β–€ ▝▀ β–˜β– β–˜β–€β–€ β–β–€β–˜β–˜ β–€β–€ Gloomberb is command-bar first. Type a ticker or a shortcut like DES AAPLor TOP and jump straight into the market view. Ranked stories, breaking news, sector feeds, global indices, FX, macro events, yield curves, movers, and sentiment.

Deutsche Bank becomes first foreign yuan clearing bank in Europe

https://tradersunion.com/uploads/images/authors/olga-shendetskaya.png
China has authorized Deutsche Bank to clear renminbi transactions in Europe, expanding the use of yuan in international trade and investment. This move strengthens a payments franchise for Deutsche Bank while broadening China's efforts to increase the currency's global role.

Choose Boring Technology (2015)

https://i.imgur.com/FRQKLCy.jpg
The author reflects on the importance of technology choices in a company, emphasizing that innovation tokens should be spent wisely and that "boring" technologies like MySQL or PHP can be good enough. They advocate for mindful choice of technology to avoid overwhelming unknowns and operational costs.

Mistral OCR 4.1

https://docs.mistral.ai/_next/image?url=%2Fassets%2Fsprites%2Fcat_idle.gif&w=640&q=75
Our latest OCR service powering our Document AI stack, with native paragraph-level bounding box extraction, structural block labels, and block-level confidence scores.

Nine PBS sues Iron Mountain over blocked access to archival data

https://current.org/wp-content/uploads/2026/08/Denver_CO_City_and_County_Building_IMG_5541-crop-scaled-e1786478301201.jpeg
Nine PBS sued Iron Mountain Data Centers for over 50 terabytes of archival materials stored in a Denver data center after their cloud-storage vendor, Open Source Storage (OSS), cut off access without warning. The station claims OSS' predecessor owned the data and Iron Mountain refuses to return it due to OSS' ownership of the infrastructure housing the data.

Ordinary abundance

https://ordinaryabundance.com/assets/abundance-room-v8.jpg
People have made significant advancements in technology and daily life, including music on demand, electric light, art reproductions, vision correction, instant communication, clean running water and refrigeration.

Understanding is the new bottleneck

https://www.geoffreylitt.com/images/talks/understanding-bottleneck/slide-01.webp?1783098284
The talk discusses the importance of human understanding in systems built by agents, and provides techniques for efficiently understanding code, including explain-diff and shared spaces. These methods aim to bridge the gap between humans and AI, enabling active participation in creative processes.

I requested a copy of my data from McDonald’s loyalty program

https://media.wired.com/photos/6a7a380554cb34d4d120f360/16:9/w_640%2Cc_limit/undefined
The author accessed their McDonald's data after signing up for the loyalty program years ago, revealing a 515-page report detailing past transactions and predicted future purchases. The report showed an alarming level of detail about the author's habits, raising concerns about commercial surveillance and data privacy.

Donkey.bas is 45 Years Old – 131 line of Glory

DONKEY.BAS shipped with early IBM PC DOS as a demo of color graphics and sound in BASICA. It was written by Microsoft co-founder Bill Gates and Neil Konzen in 1981 (version 1.10 in 1982). You only switch lanes β€” avoid the donkey, or BOOM. This page recreates the original CGA gameplay in JavaScript with some liberties. Original source: DONKEY.BAS Β· GitHub

Choosing an AI model: one prompt, 11 models, different results

https://cdn.sanity.io/images/o0o2tn5x/production/2474d03eaa7910cf4f6c5c6fa3d95fdb84ccabf3-2048x3022.png?w=208&auto=format&q=80
Netlify has launched a partnership with OpenRouter, offering two new pieces of functionality: the AI Gateway and Agent Runners. Users can now choose from various open models for their projects, including Kimi K3, GLM 5.2, and DeepSeek V4.

Kubernetes on Oxide: How customer needs shaped our integrations

https://oxide-computer-js1cdbvcf-oxidecomputer.vercel.app/img/meta/og-blog.png
The author, as Oxide's first Solutions Software Engineer, aimed to build software to solve customer problems by integrating Kubernetes with Oxide.

AI agents lie, cheat and steal. That is putting off users

NP-overrated

NP-hard problems are often considered intractable but not impossible to solve, with many practical solutions existing. In practice, the worst-case scenarios rarely occur and can be mitigated with timeouts or heuristics.

Single log line is 49KB+ (ext4) / 110KB+ (btrfs) of systemd-journald disk writes

https://private-user-images.githubusercontent.com/147394/531625159-41d31a0c-2963-4d22-8d17-73bafa87e08b.png?jwt=eyJ0eXAiOiJKV1QiLCJhbGciOiJIUzI1NiJ9.eyJpc3MiOiJnaXRodWIuY29tIiwiYXVkIjoicmF3LmdpdGh1YnVzZXJjb250ZW50LmNvbSIsImtleSI6ImtleTUiLCJleHAiOjE3ODY2NTM2MDQsIm5iZiI6MTc4NjY1MzMwNCwicGF0aCI6Ii8xNDczOTQvNTMxNjI1MTU5LTQxZDMxYTBjLTI5NjMtNGQyMi04ZDE3LTczYmFmYTg3ZTA4Yi5wbmc_WC1BbXotQWxnb3JpdGhtPUFXUzQtSE1BQy1TSEEyNTYmWC1BbXotQ3JlZGVudGlhbD1BS0lBVkNPRFlMU0E1M1BRSzRaQSUyRjIwMjYwODEzJTJGdXMtZWFzdC0xJTJGczMlMkZhd3M0X3JlcXVlc3QmWC1BbXotRGF0ZT0yMDI2MDgxM1QyMDM1MDRaJlgtQW16LUV4cGlyZXM9MzAwJlgtQW16LVNpZ25hdHVyZT1hYmRkYTlhMDAzYjkyZWE0NWUwN2E0ZmM1Mjc3M2E4N2E5ODFjMDNlM2IxNjVjZTJjMWVhNWRmNjgzMDVkYjFhJlgtQW16LVNpZ25lZEhlYWRlcnM9aG9zdCZyZXNwb25zZS1jb250ZW50LXR5cGU9aW1hZ2UlMkZwbmcifQ.NPr8Z44i9SkV4fVffjwpyhYposdeMtDrMduDRZYlPRs
Using journald in write mode on an XFS file system, it writes to the hard drive but is inefficient and can corrupt during unclean reboots. This shows that iotop's accuracy issue may be due to kernel mechanisms rather than journald itself being slow.

I built a 500k-domain search engine for makers in a weekend for $10

https://alexmorleyfinch.github.io/marlin/history/v1/article/images/hero-search.png
A user built a personal search engine over a weekend, cataloging 560,000 sites for $10 and renting an overnight GPU. The engine uses a small local language model to summarize pages and has features like fuzzy matching and category filtering.
A 17-year-old database backup of the old 0.mk URL shortener was restored and crawled, revealing that only a small percentage of links still load. The majority of links point to non-existent pages or have been blocked by websites.

DeepSeek API Pricing Update

https://pbs.twimg.com/media/HPmOJZlbUAAOgVt?format=webp&name=medium
DeepSeek-V4-Pro is launched with major Agent upgrades and flexible reasoning efforts, offering native OpenAI Responses API support. The new pricing model introduces peak and off-peak rates, which may affect users in different time zones differently.

Heart aerospace completes first flight of largest electric aircraft

https://www.heartaerospace.com/_next/image?url=%2Fapi%2Fmedia%2Ffile%2F20260812-3837%2520(1).jpg&w=3840&q=75
Heart Aerospace successfully flew its X1 demonstrator aircraft, a 106-foot battery-electric plane, for 27 minutes. The flight demonstrated all-electric flight at commercial airline scale and aims to lower air travel costs through electrification.

How Compaction Works in Pi

https://earendil.com/static/og/posts/compaction-in-pi.png
Pi uses compaction to summarize conversation history when context limit is reached, discarding older content while preserving recent work. Compaction can be manually triggered using /compact command or automatically after a turn ends with configurable token budget.

Text AI watermarks will always be trivial to remove

https://www.seangoedecke.com/og-image.jpg
The European Union's AI Act will require all AI outputs to be "detectable as artificially generated" starting in August 2026, with a new watermarking requirement.

AI At Home Part 1: A Box Of Scraps

https://jdagostino.github.io/ai-pt1-box-o-scraps/headline_server_image.jpg
The author is building a custom AI server using old computer parts, including four GPUs and an Intel Core i9 processor, to run the Deepseek V4 Flash model. They've overcome initial challenges with booting and fan control to get the system up and running with Ubuntu 24.04.

Can I use my Outputs to train an AI model?

https://downloads.intercomcdn.com/i/o/487548/17213f6a445c8e6e874b1f4b/fad85208982e639d11b9108df895a293.png
Anthropic restricts using Outputs to train AI models without permission, citing safety and competitive concerns. Customers can use Outputs for non-competitive purposes such as generating content or analyzing data.

Launch HN: Bullet (YC S26) – A Faster Coding Agent

https://www.codewithbullet.com/assets/demo.jpg
A tool called Bullet was created to speed up agent runs by eliminating unnecessary steps and parallelizing independent tasks, saving hours of waiting time. It is now used internally at the company that developed it, which aims to help others achieve similar results.

Rapid warming may tip AMOC at 2Β°C, slower warming may avert collapse

We’re checking your connection to prevent automated abuse

How AI text watermarking works

https://declaude.org/watermarking/og.png?v=2
A statistical mark is hidden in generated text by tilting the model's shortlist of words towards a secret key, making it survive copying and editing. The mark lives in the choices between words, not in the characters themselves.

How GΓΆdel's Proof Works (2020)

https://www.quantamagazine.org/wp-content/uploads/2017/04/Abstractions-520x520.png
Kurt GΓΆdel's 1931 incompleteness theorems showed that any set of axioms is either incomplete or inconsistent, proving math cannot be a complete system. This discovery has far-reaching implications in mathematics and beyond.

The Conceptual Reasoning Index

https://alignment.anthropic.com/2026/conceptual-reasoning-index/fig1.png
Researchers developed three conceptual reasoning benchmarks to evaluate AI models' ability to reason about complex issues, including decision theory and risks from advanced AI. The Conceptual Reasoning Index (CRI) aggregates these benchmarks to measure a model's overall conceptual reasoning capabilities.

We eliminated 1,400 CVEs in NanoClaw's container images

https://cdn.prod.website-files.com/680495205e33f0a1b96c7a98/6a7319dce7da1c3c4d7e6e71_image%20(9).png
Echo scans the NanoClaw container with multiple independent scanners to identify vulnerabilities, then separates findings into "safe to bump" and "needs real work." Echo's proprietary backporter agent applies security patches from newer versions of packages to older versions.

How Organizations Use AI: Evidence from ChatGPT [pdf]

Researchers studied how organizations use generative AI, specifically ChatGPT Enterprise, by linking account records to usage data from March 2026. The study found that ChatGPT Enterprise usage has grown rapidly among large and valuable firms with high R&D and SG&A expenses.

Solid 2.0 RC: The Big <Reveal>

Graduate student proves a quantum uncertainty principle for fractals

https://www.quantamagazine.org/wp-content/uploads/2026/08/Fractal-Uncertainty-cr-Ada-Zejun-Shen-lede-scaled.webp
A mathematician proved a new uncertainty principle that applies to fractals, which are shapes with intricate patterns. This principle shows that quantum particles cannot be trapped in fractal-like paths forever.

Tocharian Online

Tocharian is a closely related Indo-European language family, consisting of two distinct languages A and B, discovered in Chinese Turkestan documents along the Silk Road. Tocharian texts date back to the 6th-8th centuries AD and include translations of Buddhist texts.

My Rules for Using Spreadsheets

https://leancrew.com/all-this/images2026/20260731-Cubic%20sum%20spreadsheet.png
The author shares their "Don't" rule for using spreadsheets, which is to avoid them for large or complex data sets due to poor organization and errors. They explain how years of working with poorly organized Excel spreadsheets soured them on the use of spreadsheets.

Show HN: MCP Memory – Fast Agent Memory Using Google's OKF and SQLite FTS5

https://opengraph.githubassets.com/67d3d1f3293a42164822fbfd740dd863381ec6688f4e7d1d547c86cbadfb1b44/fellowgeek/mcp-memory
MCP-Memory is a server for AI agents with persistent memory capabilities using the Open Knowledge Format (OKF v0.2) standard and SQLite indexing. It enables long-term memory storage, search, and retrieval across sessions with automatic setup and configuration.

AI Is Threatening Natural Resources for Billions

https://e4k4c4x9.delivery.rocketcdn.me/en/wp-content/uploads/sites/15/2026/06/Data-center-696x418.jpg
By 2030, AI's water use will match the needs of 1.3 billion people and its power use triples that of 650 million, according to a UN University report.

GoAccess – Open-source real-time log analyzer and interactive viewer

https://goaccess.io/images/goaccess-dashboard-2026-2.png
GoAccess is a fast, terminal-based log analyzer for real-time web server statistics. It detects suspicious activity and generates reports in HTML, JSON, and CSV formats.

PBS loses 70 years of TV history after cloud storage vendor goes defunct

https://cdn.mos.cms.futurecdn.net/xi79WuWDZXzix4Fc7sXNMn.png
Nine PBS is facing a crisis after its cloud storage vendor, Open Source Storage (OSS), went out of business and lost access to over 50TB of data. The channel is now suing OSS and Iron Mountain, which it believes holds the data, to ensure its preservation.

US sells 30-year bonds at highest borrowing costs since 2001

https://barrier-page-components.s3.eu-west-1.amazonaws.com/assets/images/why-ft-text.svg
Then $75 per month. Complete digital access to quality FT journalism on any device. Cancel anytime during your trial. Essential digital access to quality FT journalism on any device. Pay a year upfront and save 20%. Complete digital access to quality FT journalism with expert analysis from industry leaders. Pay a year upfront and save 20%.

Gemini 3.7 Flash

https://storage.googleapis.com/gweb-uniblog-publish-prod/images/gemini-3-7-flash__evals__perform.width-1200.format-webp.webp
Gemini 3.7 Flash is the latest model, offering substantial improvements in coding and agent tasks with a lower price point than Gemini 3.6 Flash. It delivers strong gains in debugging, issue resolution, and code accuracy across various workflows and domains.

Time to Move On: Querying Without Nulls and Bags

https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png
The SQL language's success stems from its declarativeness, but it tolerates bags and nulls that hinder optimal performance. A null-free language with set semantics can avoid these issues and provide better results.

JDK 27 G1/Parallel/Serial GC Changes

OpenJDK 28 is approaching with significant changes to its stop-the-world collectors, including the introduction of G1 as the default garbage collector. These updates aim to improve performance, footprint, and latency profile by addressing various issues and adding new features.

Flock updates privacy, accountability, security, and transparency safeguards

https://cdn.prod.website-files.com/6823a65a40e8919f48f55a9d/6a7d007b414f5a757dc1f8d0_Untitled%20design.jpg
Flock is updating its ALPR data retention period to 7 days and introducing Evidence Mode for preserving specific data during active investigations. The company also strengthens oversight of system use with new features such as offense filtering, required audit assistance, proactive lockouts, and case codes.

Mozilla says stricter antitrust measures against Google 'threaten Firefox'

https://www.techcentral.ie/wp-content/uploads/2017/03/Firefox_FF-e1786526594888.jpg
Mozilla may exit the browser market if it can't receive payments from Google for offering its search engine. This could lead to a loss of privacy-focused browsers and increased power for large tech companies.

Hello, me. It's been a while

The author reflects on how they've lost the habit of quiet thinking over 14 years, replaced by constant background noise. They rediscovered this habit and found it to be more enjoyable than filling silence with podcasts or social media.

I want extern "fil-C"

The author proposes a bridge between Rust and C, Fil-C, which compiles C with capabilities and runtime checks for memory safety. This would allow existing software to be rewritten in Rust while maintaining performance on legacy code that remains safe under ordinary unsafe C.

Build a Stratum 1 PTP Grandmaster on a Budget

https://opscode.io/images/ptp-topology-router.png
A DIY project built a PTP (Precision Time Protocol) grandmaster clock using a Raspberry Pi CM4, an Intel i226 NIC, and a GPS module. The build achieved sub-microsecond synchronization accuracy over LAN and WAN networks.

McDonald's Built a 515-Page Dossier on Me. It Says I'll Never Stop Eating There

https://media.wired.com/photos/6a7a380554cb34d4d120f360/16:9/w_640%2Cc_limit/undefined
The author accessed their McDonald's data after signing up for the loyalty program years ago, revealing a 515-page report detailing past transactions and predicted future purchases. The report showed an alarming level of detail about the author's habits, raising concerns about commercial surveillance and data privacy.

DeepSeek up to 1000% price hike is live

https://pbs.twimg.com/media/HPmOJZlbUAAOgVt.jpg?name=small&format=webp
API pricing update πŸ’° With the V4 lineup release, we’re updating our API pricing and introducing peak and off-peak rates. Off-peak rates are 50% lower than peak, enabling more flexible workload scheduling. πŸ“‰ New pricing takes effect at 16:00 UTC, Aug 16, 2026 πŸ•’

HN: Xenon, open source dashboard for a second screen, tablet or phone

https://xenon-app.com/images/xenon-ai-chat.png
Pull the top bar down and type the way you would say it, like "photos from december". The same index, kept current on the PC in the background, is what draws the map of a drive the moment you open it; cleanup proposes only what is on a fixed list of caches and temporary files, and what it removes goes to the Recycle Bin.

Blog about things you don't understand yet

https://www.seangoedecke.com/og-image.jpg
The author publishes blog posts that represent at least two things they've learned, forcing them to think critically and research topics thoroughly. This process helps clarify their thoughts and leads to more effective writing.

Airbnb Eval-driven development: Lessons from evaluating GenAI at scale